期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2015
卷号：80
期号：2
出版社：Journal of Theoretical and Applied
摘要：The main focus from one of the problems in computer networks is computer security systems because of the high threat of attack from the internet in recent years. Therefore, an Intrusion Detection System (IDS) that monitors the traffic of computer networks and oversight of suspicious activities in a computer network is required. Research on intrusion detection system have been carried out. Several researches have used artificial neural networks combined with a fuzzy clustering method to detect attacks. However, there is an issue that arise from the use of such algorithms. A single artificial neural network can produce overfitting on intrusion detection system output. This research used two methods of artificial neural networks, namely Lavenberg-Marquardt and Quasi-Newton to overcome that issue. Both algorithms are used to detect computer networks from attack. In addition, the use Possibilistic Fuzzy C-Means (PFCM) before going into the neural network ensemble with simple average. Then on the output, Naive Bayesian classification method is used. Dataset used in the research were NSL-KDD dataset which is an improvement of KDD Cup'99. KDDTrain+ used for training data and KDDTest+ for testing data. Evaluation results show good precision in detection of DoS (89.82%), R2L (75.78%), normal (72.25%) and Probe (70.70%). However, U2R just get 14.62%. At recall, good results achieved by normal state (91.44%), Probe (87.11%) and DoS (83.31%). Low results occurred in U2R (9.50%) and R2L (6.14%). Meanwhile, lowest accuracy on normal category (81.18%) and highest in U2R (98.70%). The results showed that the neural network ensemble method produces a better average accuracy than previous researches, amounting to 90.85%.
